This study answers fundamental questions of large-scale neural networks in the human brain supporting crossmodal cognition. To reveal how auditory and visual stimuli and motor acts are arbitrarily combined as a result of crossmodal learning and integrated to supramodal symbolic representations, we will study the neural representations of the letters of the Roman alphabet. These consist of four unimodal representations (visual, auditory, and motor representations for writing and speaking) and learned connections between these, that is, the processes that underlie their audiovisual recognition and motor production. Accurate experimental control is facilitated by the fact that letters exhibit all the necessary properties of symbolic crossmodal representations but in a physically simple and exact format carrying no semantic associations that could confound the neurophysiological interpretation of results. Combined 3-Tesla functional magnetic resonance imaging (fMRI) and 306-channel magnetoencephalographic / 128-channel electroencephalographic (MEG/EEG) techniques with simultaneous behavioral recordings will be applied to pinpoint the exact underlying neural mechanisms. This approach combines the advantages of spatially accurate fMRI with temporally specific MEG/EEG, enabling accurate spatiotemproal characterization of brain activity totally noninvasively. To directly observe how large-scale neurocognitive networks evolve during crossmodal associative learning, we will also conduct fMRI/MEG/EEG measurements before and after our subjects are taught (previously unfamiliar) Japanese kana-letters. The specific aims are to elucidate structure, function, and oscillatory mechanisms of fully established crossmodal neural networks based on previous extensive associative learning (Roman letters) and currently evolving networks representing novel crossmodal associations (Japanese letters). We will characterize the relative roles of deep brain nuclei, cerebellum, medial temporal lobe, and sensory-specific and multisensory association cortices in such networks. The multidimensional experimental design allows isolation of neural mechanisms utilized by perception, working memory, memory encoding, and recall.